Incidence of incorrect annotation of radiological images is common in radiology department and can lead to grave consequence of patient. At radiology department, image annotation about the laterality (e.g. Left, Right) or view position (e.g. AP, PA) is required for every image. As the annotation process is a pure human performed procedure, human error is inevitable. The common type of incorrect image annotation includes wrong laterality of extremity images and falsely flipped Chest images. Falsely flipped Chest images can cause the patient misdiagnosed as dextrocardia. If physician based on the falsely flipped Chest image to perform the drainage procedure for pleural effusion, a wrong side chest drain incidence can occur. Wrong laterality of extremity images is the other commonly occurred incidence at radiology department. For example a left hand X-ray image may be labeled with right, such error can potentially mislead radiologist and result in wrong side report. In order to minimize the incorrect annotation incidence at radiology department of Pok Oi Hospital, a radiological quality control application SureSide is developed to minimize the wrong annotation of radiological images.
SureSide is a self developed DICOM application with computer vision capability to analyze the dicom tag and the annotation of radiological image, it is a server-client application with server situated at PACS network to receive auto-routed DICOM objects from PACS server and client installed at the workstations of radiological exam room to display the warning windows if incorrect annotation is detected at the image. By extracting the pixel data of the DICOM object, the image is evaluated by the open source computer vision library OpenCV to detect image of digital marker or physical maker. Simple computer vision technique “template matching” is used to detect image of digital marker, while feature descriptor named “KAZE” is used to detect the image of physical marker. If discrepancy is found for image’s dicom tag of “View Position” and the image of annotation, a warning window will pop up at SureSide’s client to alert Radiographer, so Radiographer can rectify the problem image as soon as possible.
Apart from the falsely flipped Chest images, SureSide is also capable to detect the wrong laterality labelled images. By counter checking the laterality of requested exam and the marker(Left/Right) applied at the image, SureSide can spot out the wrong laterality marker.
Co-location with:
